Agent skills

MCP tools expose what Glyphs MCP can do. Agent skills describe how an agent should use those tools for a specific job.

This matters because type-design work is contextual. The correct action often depends on the current font, master, selection, metrics keys, kerning groups, or selected nodes. Skills encode the repeatable habits that keep those sessions predictable.

Why skills help

Skills help agents:

• prefer non-mutating review tools before any edit tool
• read context before acting
• prefer dedicated tools over free-form code
• use dry runs before mutation
• avoid auto-saving
• fetch only relevant bundled docs through docs_search and docs_get

Current skill bundle

Skill	Use it for
glyphs	General Glyphs MCP tasks, context checks, and routing to focused workflows.
glyphs-mcp-development	Reusable workspace Python scripts and six plug-in types.
glyphs-mcp-outlines-docs	Outline work and its dedicated live-code fallback, selected nodes, components, anchors, and focused docs lookup.
glyphs-mcp-italic-first-pass	Experimental Roman-to-italic construction drafts and their narrowly guarded Python fallback.
glyphs-mcp-scripting	Safe live vibe coding, Macro Panel snippets, and iterative Python debugging in the running app.
glyphs-mcp-features	OpenType features and stylistic-set glyph groups.
glyphs-mcp-icon-font	Stable Unicode and PUA assignment workflows for icon and symbol fonts.
glyphs-mcp-kerning	Kerning bumper review and approved apply steps.
glyphs-mcp-litsquare-metadata	LitSquare metadata inspection, inheritance, safe patches, semantic path roles, and fixed layer-specific IconGrid centering.
glyphs-mcp-spacing	Spacing review, dry-run, and approved apply steps.
glyphs-mcp-release	Version, documentation, package synchronization, release validation, signing, and publication gates.

The glyphs skill is a small general launcher. The remaining skills stay intentionally narrow and provide operating instructions for common focused workflows.

Source of truth

Authored skills live in skills/.

The repo exposes them to clients through:

• .agents/skills for Codex, Cursor, and GitHub Copilot CLI
• .claude/skills for Claude Code

The Codex and Claude Code installers can copy the same skill bundle into their global skill directories. Cursor can use the repo-local .agents/skills bridge, an explicit .cursor/skills project copy, or a manual ~/.cursor/skills installation. GitHub Copilot CLI can use the same repo-local .agents/skills bridge when its optional plugin is not installed.

The optional Codex/ChatGPT, Claude Code, Cursor, and GitHub Copilot plugins all carry the same synchronized skills/ directory and .mcp.json connection. scripts/sync_codex_plugin_skills.sh keeps its historical filename, but now manages the shared agent package: it copies the eleven canonical directories from skills/ into plugins/glyphs-mcp/skills/. Use its --check mode in release validation, and do not edit packaged copies independently.

Each host manifest shares the Glyphs MCP release version. Skills inherit that package version instead of maintaining eleven independent version fields, while the running MCP server reports the native Glyphs MCP plug-in version.

When to invoke a skill

Name the skill when you want the agent to follow the workflow:

Use the glyphs-mcp-spacing skill to review spacing for the selected glyphs and do a dry run first.

Continue with Use skills for setup details.